pytorch/xla

Pull Requests Commits

Some command line args for run_tests.sh

dlibenzi committed 7 years ago

75a01fa2

Make flags clang dependent.

dlibenzi committed 7 years ago

30f1dc7b

Update the PyTorch commit id to latest

Alex Şuhan committed 7 years ago

07ff8dd7

Make addmm parameters const refs

Alex Şuhan committed 7 years ago

6fced970

Move big tensor method implementations to helpers

Alex Şuhan committed 7 years ago

4c97a0a7

[XLATensor] Add addmm

Alex Şuhan committed 7 years ago

6c8298dc

[XLATensor] Add avg_pool2d

Alex Şuhan committed 7 years ago

d17b883a

[XLATensor] Add size method

Alex Şuhan committed 7 years ago

f501d7eb

Expand tensor type mapping to the full list of supported tensor types.

dlibenzi committed 7 years ago

45f42939

Some more code reuse.

dlibenzi committed 7 years ago

ae0e00d4

dlibenzi committed 7 years ago

9adbcecd

Some more code refactoring.

dlibenzi committed 7 years ago

c63ed9b7

More refactoring.

dlibenzi committed 7 years ago

83f6611f

Further optimizations and cleanups.

dlibenzi committed 7 years ago

a89df65a

More refinements to the tensor interface.

dlibenzi committed 7 years ago

7969b4f8

Added more fine grained support for PT<->XLA integer types.

dlibenzi committed 7 years ago

88c908a7

Further tensor util code simplification.

dlibenzi committed 7 years ago

b823e80d

Be consistent in the IR Node ToString value separation.

dlibenzi committed 7 years ago

1c6c3bb8

Added C++ test utility to test on all available device types.

dlibenzi committed 7 years ago

73a3f36c

Added C++ unittest util file and added namespaces.

dlibenzi committed 7 years ago

728540bd

Redirect C++ test log to file instead of suppressing it.

dlibenzi committed 7 years ago

84266bb4

Made C++ tensor test work and removed logging when testing.

dlibenzi committed 7 years ago

e07d5d4d

Added simple tensor test (missing check as ATEN throws) and minor refactoring of XLA tensor.

dlibenzi committed 7 years ago

86d684c8

Made literal to tensor conversion use CopyTensors, which is case or relayout is considerably faster.

dlibenzi committed 7 years ago

9668dd45

dlibenzi committed 7 years ago

70ef4e84

Avoid literal-to-tftensor copy by populating the tftensor buffer directly.

dlibenzi committed 7 years ago

736a75c4

XLA tensor restructuring. Moved conversion functions into a util facility, and Device into dedicated file.

dlibenzi committed 7 years ago

7d295e2f

Made XLATensor references back to constant.

dlibenzi committed 7 years ago

1adef744

Finishing up handling the 3way nature that XLA tensors can have (XLA data, IR Node, or ATEN tensor).

dlibenzi committed 7 years ago

c5625737

[XLATensor] Add threshold method

Alex Şuhan committed 7 years ago

e217d826

Older